introduction to risk factors & measures of effect meg mccarron, cdc
TRANSCRIPT
Introduction to Risk Factors & Measures of Effect
Meg McCarron, CDC
2
Introduction to
Risk Analysis
What is a risk analysis?• The analysis of an association between a variable (e.g.
underlying condition) and an outcome (e.g. death)
• Why do risk analysis?
• The probability of an outcome is often dependent on the interplay between a variety of factors
• Follow up on suggested associations observed in descriptive analysis (e.g. the elderly appear to die more frequently than healthy young adults; a risk analysis might tell you whether or not that is a true observation)
• Determine the severity of risk
• Identify significant risk factors
• Using this type of analysis we can measure risk ratio (RR), odds ratio (OR) 3
What is a risk factor?
A risk factor is a factor that is associated with increased chance of getting a disease.
In epidemiological terms: A risk factor is a variable (determinant) associated with an increased risk of disease or infection (outcome).
Example: Obesity (determinant/exposure) is associated with increased risk of heart attack (outcome)
When we measure risk factors we assess Strength Direction Shape
4
Risk factors in SARI surveillance
• Information about a number of potential risk factors and outcomes is often recorded• e.g. Outcomes: death, influenza status• Risk factors: age, co-morbid conditions
• Surveillance data can be analyzed to increase the understanding of the association of risk factors with severe outcomes
• Surveillance data describing exposures allows analysis of associations without expensive in-depth studies
5
Is a risk factor the cause of a disease?
Risk factors are correlational and not necessarily causal Correlation does not imply causation The statistical methods used do not consider
the direction of effects For an effect to be causal the exposure must
have occurred before the outcome e.g. young age does not cause measles
(Morbillivirus causes measles), but young people are at greater risk because they are less likely to have developed immunity due to previous exposure or vaccination
6
The Correlation-Causation Problem
Somalia has many pirates, but low carbon emissions
How are risk factors/disease determinants identified?
Individual-level data Two key variables
Outcome: e.g. influenza Exposure: e.g. vaccination
Should consider multiple risk factors
Epidemiological study designs used to identify risk factors Case-control Cohort
Surveillance data may approximate a cohort study
Biological plausibility e.g. age and influenza infection Exposure (risk factor) must occur prior to outcome
(disease)
Types of variables
Continuous E.g. Age
Categorical variables Binary
E.g. Gender, vaccination status Ordinal
E.g. Age group, socioeconomic status (SES) Nominal/Categorical
E.g. Geographic region Count
E.g. number of ILI symptoms
How are risk factors/disease determinants identified?
Clinical and epidemiological comparison of hospitalized SARI patients with and without laboratory-confirmed influenza week 40/20xx to (current week)/20xx, Country X (NOTE: Numbers in table are not real and for example only) Characteristics Percent of influenza-negative SARI
hospitalizations with selected demographic and epidemiological characteristics
Percent of SARI hospitalizations confirmed as influenza with selected demographic and epidemiological characteristics
Sex Information available for N = 100 Information available for N = 50
Male 54/100 (54%) 27/50 (54%) Female 46/100 (46%) 23/50 (46%) Sex unknown 0 0
Chronic Medical Illnesses Information available for N = 98 Information available for N = 48 Number of cases with at least one of the
chronic medical illness listed below *
30/98 (31%) 28/48 (58%)
Chronic respiratory disease 15/98 (15%) 20/48 (42%) Asthma 15/98 (15%) 10/48 (21%) Diabetes 11/98 (11%) 11/48 (23%) Chronic cardiac disease 5/98 (5%) 5/48 (10%) Chronic renal disease 3/98 (3%) 3/48 (6%) Chronic liver disease 4/98 (4%) 4/48 (8%) Chronic neurological impairment 7/98 (7%) 6/48 (13%) Immune-compromised 0/98 (0%) 1/48 (2%) Number of cases without any of the above chronic medical illnesses
68/98 (69%) 20/48 (42%)
Unknown if risk factors present N=2 N=2 Pregnancy status Information available for N = 50 women Information available for N = 23 women
Pregnancy in any trimester 11/50 (22%) 8/23 (35%) Not-pregnant 39/50 (78%) 15/23 (65%) Pregnancy status unknown N=0 N=0
Obesity (or other conditions as determined by national priorities)
Information available for N = 90 Information available for N = 35
Obese (BMI>30 or judged obese clinically) 25/90 (28%) 15/35 (42%) Not obese (BMI<30 or not clinically judged obese) 65/90 (72%) 20/35 (58%) Obesity status unknown 10 15
Age-groups (years) Information available for N = 100 Information available for N = 48 0-1 40/100(40%) 10/48 (21%) 2-4 25/100(25%) 8/48 (17%) 5-14 10/100 (10%) 10/48 (21%) 15-29 5/100 (5%) 11/48 (23%) 30-64 5/100 (5%) 8/48 (16%) 65+ 15/100 (15%) 1/48 (2%) Age unknown N=0 N=2
Vaccination Status Information available for N = 98 Information available for N = 40 Received monovalent or trivalent vaccine during the current influenza season
40/98 (41%) 2/40 (5%)
Did not receive monovalent or trivalent vaccine during the current influenza season
58/98 (59%) 38/40 (95%)
Vaccination status unknown N=2 N=10 Oseltamivir/zanamivir (Tamiflu/Relenza) Use Information available for N = 100 Information available for N = 44 Received oseltamivir/zanamivir within 48 hours of symptom onset
10/100 (10%) 8/44 (18%)
Did not receive oseltamivir/zanamivir within 48 hours of symptom onset
90/100 (90%) 36/44 (82%)
Oseltamvir use unknown N=0 N=6 Median days from symptom onset to hospital admission 4.0 days 4.5 days
10
How are risk factors/disease determinants identified? (… continue …)
Clinical and epidemiological description of hospitalized SARI patients with laboratory-confirmed influenza, by outcome status, year x to year y, Country X (NOTE: Numbers in table are not real and for example only)
Hospitalised SARI cases with laboratory-confirmed influenza Characteristics
Percent of hospitalized (non-ICU/non-severe) cases with selected demographic and epidemiological characteristics
Percent of severe (severe outcome/or died) cases with selected demographic and epidemiological characteristics
Sex Information available for N = 100 Information available for N = 30
Male 54/100 (54%) 15/30 (50%) Female 46/100 (46%) 15/30 (50%) Sex unknown 0 0
Chronic Medical Illnesses Information available for N = 98 Information available for N = 28 Number of cases with at least one of the chronic medical
illness listed below *
30/98 (31%) 19/28 (58%)
Chronic respiratory disease 25/98 (25%) 20/28 (71%) Asthma 15/98 (15%) 4/28 (14%) Diabetes 11/98 (11%) 54/28 (23%) Chronic cardiac disease 5/98 (5%) 5/28 (18%) Chronic renal disease 3/98 (3%) 3/28 (11%) Chronic liver disease 0/98 (0%) 4/28 (14%) Chronic neurological impairment 3/98 (3%) 7/28 (25%) Immune-compromised 0/98 (0%) 1/28 (4%) Number of cases without any of the above chronic medical illnesses
68/98 (69%) 9/28 (42%)
Unknown if risk factors present N=2 N=2 Pregnancy status Information available for N = 50 women Information available for N = 15 women
Pregnancy in any trimester 11/50 (22%) 10/15(67%) Not-pregnant 39/50 (78%) 5/15 (33%) Pregnancy status unknown N=0 N=0
Obesity (or other conditions as determined by national priorities)
Information available for N = 90 Information available for N = 28
Obese (BMI>30 or judged obese clinically) 23/90 (26%) 19/28 (68%) Not obese (BMI<30 or not clinically judged obese) 66/90 (73%) 9/28 (32%) Obesity status unknown 10 2
Age-groups (years) Information available for N = 100 Information available for N = 30 0-1 35/100(35%) 5/30 (17%) 2-4 30/100(30%) 2/30 (6%) 5-14 10/100 (10%) 5/30 (17%) 15-29 4/100 (4%) 3/30 (10%) 30-64 6/100 (6%) 10/30 (33%) 65+ 15/100 (15%) 5/30 (17%) Age unknown N=0 N=0
Vaccination Status Information available for N = 98 Information available for N = 30 Received monovalent or trivalent vaccine during the current influenza season
20/98 (20%) 2/30 (7%)
Did not receive monovalent or trivalent vaccine during the current influenza season
78/98 (80%) 28/30 (93%)
Vaccination status unknown N=2 N=0 Oseltamivir/zanamivir (Tamiflu/Relenza) Use Information available for N = 100 Information available for N = 27 Received oseltamivir/zanamivir within 48 hours of symptom onset
15/100 (15%) 2/27 (7%)
Did not receive oseltamivir/zanamivir within 48 hours of symptom onset
85/100 (85%) 25/27 (93%)
Oseltamvir use unknown N=0 N=3 Median days from symptom onset to hospital admission 3.5 days 7.5 days
11
Cohort study Follow people over
time Collect data on
their exposures (risks)
Monitor their outcomes
Compare risk of disease among exposed versus unexposed
Participant
1
2D
3
4D
5
6
0 1 2 3 4
time
Example: cohort study
e.g. Risk of death among SARI admissions Outcome: death Risk factors: age, underlying conditions,
influenza-positive Source population: all patients admitted
with SARI, followed until death or discharge
13
14
Case control study Cases: people with
disease Deliberately over-
selected Controls: people without
disease Represent exposure
distribution of the source population
Find out their exposure status
Compare risk of exposure among diseased and non-diseased
E D 1
Participant
D 2
E D 3
4
E 5
6
time
Example: case-control study
Risk of influenza among vaccinated patients Cases: people with influenza Controls: people without influenza Outcome: influenza status Risk factors: vaccination status, age,
underlying comorbidity
15
Statistical significance: is the association due to chance alone?
A statistical test is used to assess if an association may be due to chance alone (random error) In statistics, a result is called statistically
significant if it is unlikely to have occurred by chance alone, according to a pre-determined threshold probability, the significance level (e.g. α: 0.05).
16
Common statistical tests
Categorical data: Chi-square (2) test, Fisher’s test McNemar’s test
Continuous data: T-test Wilcoxon rank-sum test ANOVA
These tests can tell if there’s a difference between groups but do not convey the size or direction of effects
Common measures of association / effect
Measure the size of an association (effect) Compare some measure of disease in exposed versus
unexposed Absolute difference
Y1-Y2
Risk difference Relative difference (ratio)
Y1/Y2
Odds ratio Risk ratio Incidence rate ratio Hazard ratio (survival data) Attributable risk
18
19
Odds ratios Most common measure
of association used in epidemiology
Binary outcome Odds Ratios (OR):
compares the odds of exposure among cases (people with disease) with controls (people without disease) Odds: ratio of the
probability (p) of an event occurring versus it not occurring
Odds = p/(1-p)
Calculation of the RR & ORCases Controls
Exposed a b
Unexposed
c d
OR = (a/c) / (b/d)
OR = 1 = no associationOR < 1 = negative association
(reduces risk)OR > 1 = positive association
(increases risk)
20
Example of OR CalculationsOutcome (Influenza patients that died)
Outcome (Influenza patients that died)
Calculation of the RR & ORDied Alive
Flu+ 200 (a) 150 (b)
Flu- 50 (c) 100 (d)
OR = (a/c) / (b/d) = (a*d) / (b*c)
OR=(200/50)/(150/100)=2.7
Calculation of the RR & ORDied Alive
Female
200 (a) 180 (b)
Male 98 (c) 100 (d)
OR=(200*100)/(180*98)=1.1
Confidence intervals OR is a point estimate Confidence interval (CI)
is a measure of uncertainty around your point estimate
CI is based on the standard error (SE)
SE=narrower confidence interval
If CI includes 1, then not statistically significant wide CI also a problem
Usually use 95%CI
Cases Controls
Exposed a b
Unexposed
c d
SE = √1/a + 1/b + 1/c + 1/d
95%CI = e(OR 1.96 * SE)
22
• OR=1.1• 95%CI=1.01,1.4
Confidence intervals
e.g. 2007 Victorian surveillance data, adults, influenza B
Flu+ Flu-
Vaccinated 44 (a) 95 (b)
Unvaccinated
205 (c) 260 (d)
OR = (44/205) / (95/260) = 0.59
ln(OR) = ln(0.25) = -0.53
SE = √1/44 + 1/95+ 1/205 + 1/260 = 0.20
95%CI = e(-0.53 + 1.96*0.20) = e(0.09) = 0.39 (UL)= e(-0.53 - 1.96*0.20) = e(-2.87) = 0.88
(LL)
Interpreting Results
Size of the CI is an indicator of uncertainty Wide CI = uncertainty Narrow CI = uncertainty
If CI includes 1, then not statistically significant The observed effect could just be due to chance
P-values are often used to convey statistical significance The p-value for a OR is calculated from a chi-
squared test The p-value reference for a 95%CI is 5% or 0.05
P-values
The p-values help us to determine whether the difference between the two groups might be due to random variation
CI and p-values 95%CI=1.0, 2.3 indicates that the two-sided p-value
for no association is about 0.05. 95%CI=0.9, 2.4 suggests p>0.05 95%CI=0.9, 2.4 indicate that the data are compatible
with a two-fold higher risk (i.e. upper limit includes 2) The p-value is a measure of the compatibility of
the data and the null hypothesis
Implementation of a statistical test
We start with a research hypothesis State the relevant null (H0)
No effect (effect is due to chance) Alternative hypotheses (HA)
An effect exists Decide which test is appropriate (see earlier list) Compute the test statistic and the associated p-
(probability) value Compare the computed p-value to a reference p value
(usually 0.05) to accept or reject the null hypothesis If the p-value of the test is lower than the reference
value the H0 is rejected The effect is not likely to be due to chance
Example: Implementation of a statistical test
Influenza prevalence in hospitalized patients:
Non pregnant women: 100/1000 = 10%
Pregnant women: 30/200 = 15%
Question: Is the influenza
prevalence in hospitalized pregnant women different to non-pregnant women?
Hypothesis H0: p1 = p2 ; p1 - p2
= 0 HA: p1 = p2 ; p1 - p2
= 0 Reject H0 if p (test) is
< α: 0.05 Test results:
Z (test statistic): 0.119 p value: 0.037
0.037<0.05 → Reject H0
Example: factors associated with influenza-positive diagnosis among ILI patients
OR p-
value
95% CI
Lower limit
Upper limit
Vaccinated 0.54 0.02 0.32 0.89Underlying condition 1.20 0.47 0.72 2.00
Epi week 1.04 0.01 1.01 1.08
Age group
<20 ref 20-64 0.76 0.17 0.51 1.1365+ 1.09 0.85 0.45 2.62
Adjusted OR=0.54 (95%CI=0.32,0.89)
Crude OR=0.59 (95%CI=0.39,0.88)
Summary
A risk factor is a variable which increases (or decreases) the risk of an outcome
We can assess the influence of risk factors using individual-level data from case-control and cohort studies
The size of the effect can be measured by effect measures Most common effect measure is the odds ratio
The uncertainty of the effect can be measured by the confidence interval Understanding whether an effect is due to random error is
indicated by the p-value and tested using a statistical test Multivariable methods can tell us how much influence
one risk factor has compared with others